HPE–Oracle Gigawatt AI Cloud Deal Shows Network Spend Rising to 24% of Infrastructure

HPE networking revenue surges as Oracle scales AI cloud

Hewlett Packard Enterprise posted $2.9 billion in Q3 2026 networking revenue, up 74.9% year over year, driven by an expanded partnership with Oracle to build gigawatt-scale GPU superclusters for Oracle Cloud Infrastructure. The partnership positions OCI as a credible alternative to AWS, Azure, and Google Cloud for GPU-intensive workloads — and signals that network infrastructure is consuming a sharply rising share of AI cloud budgets.

HPE's overall Q3 revenue hit $12.2 billion, up 34% year over year, while Oracle reported $19.2 billion in Q4 2026 revenue, up 21%, with full-year revenue reaching $67.4 billion. To align incentives, HPE issued stock warrants to Oracle as part of the deal structure, underscoring the strategic weight both companies place on the partnership.

For enterprise buyers, the shift is clear: network spend is no longer a rounding error in AI infrastructure. HPE's 75% networking growth rate — against a 34% overall revenue increase — shows that high-speed switching, routing, and optics are taking budget share from servers and storage. Enterprises planning GPU cluster deployments should allocate proportionally more capital to 400G/800G fabric upgrades, InfiniBand alternatives, and low-latency routing rather than assuming network costs scale linearly with compute.

Multi-cloud leverage improves as OCI backs gigawatt AI capacity

OCI's commitment to multi-gigawatt data center build-outs using HPE Juniper Networking creates a fourth credible hyperscale option for AI workloads, alongside AWS, Azure, and Google Cloud. This matters for enterprise multi-cloud bargaining power: buyers with GPU-intensive workloads now have a viable alternative backed by HPE's hardware roadmap, not just incremental capacity from a second-tier provider.

The partnership targets the same AI infrastructure layer where Microsoft, AWS, and Google are building GPU superclusters, and where Cisco, Arista Networks, and NVIDIA's networking stack (Mellanox/InfiniBand) compete for fabric dominance. HPE's Juniper acquisition — now showing 75% revenue growth in this segment — positions it to take share from Cisco and Arista in cloud-scale AI fabrics.

Buyers with hybrid Oracle–HPE footprints can justify standardizing on Juniper-based routing and switching for high-performance workloads, simplifying reference architectures and reducing multi-vendor support overhead. However, this deepens cross-vendor dependency risk: large buyers should explicitly model joint HPE–Oracle outage scenarios and supply-chain disruptions in cloud risk registers, particularly for workloads pinned to OCI regions.

Anthropic's $517B compute contracts reshape hyperscaler capacity allocation

Reporting by The Information indicates that Anthropic has signed compute contracts worth up to $517 billion over eleven months, locking in at least 14.8 gigawatts of infrastructure capacity. These multi-year deals — likely spanning AWS Bedrock and Google Cloud partnerships — turn foundation model companies into anchor tenants that materially shape hyperscaler infrastructure planning.

When hyperscalers commit double-digit gigawatts to a handful of AI anchor tenants, enterprises face capacity constraints and pricing pressure for high-end GPU instances during peak demand. The contract structure also sets a benchmark: enterprises spending hundreds of millions annually on cloud can push for reserved GPU capacity with explicit gigawatt or rack-level commitments, plus long-term price curves instead of volatile spot pricing.

The risk for enterprises is priority scheduling: anchor tenants like Anthropic may get precedence in GPU allocation during shortages, leaving enterprise buyers with longer lead times or premium spot pricing. Multi-cloud GPU strategies reduce this exposure, but require consistent workload portability across AWS, Azure, Google Cloud, and now OCI.

$4.07B in data center funding concentrates in AI capacity plays

Twelve documented data center financings in September 2026 totalled $4.07 billion, with 83% concentrated in the four largest rounds. Named deals include Starcloud's $250 million Series A extension for orbital data center facilities, plus at least $3.75 billion across Firmus, Volta Infra, Fluidstack, Armada, and Panthalassa for physical AI-optimized capacity.

This capital concentration signals that AI infrastructure is attracting large-scale, multi-year private investment, not just hyperscaler capex. For enterprise buyers, it means more capacity options outside AWS/Azure/Google, particularly for workloads that can tolerate edge or geographically novel deployments (e.g., maritime, satellite-adjacent, or remote regions). However, these providers carry higher vendor risk than established hyperscalers — enterprises should require explicit SLA guarantees, escrow arrangements, and financial stability disclosures before committing production workloads.

What to watch

Network cost as a budget line item: If HPE's 75% networking growth holds across the industry, enterprises should expect network infrastructure to rise from ~15% to 20–25% of total AI cloud spending over the next 12 months. Budget models that assume fixed network-to-compute ratios will underestimate total cost of ownership.

Hyperscaler contract structures: Anthropic's reported $517B commitment shows hyperscalers will sign long-duration, capacity-secured deals with anchor tenants. Enterprises with large cloud footprints should test whether their account teams can offer similar structures — reserved gigawatt-level capacity with fixed pricing — particularly for GPU-intensive workloads.

Vendor concentration risk in OCI–HPE deployments: Buyers adopting OCI for AI workloads inherit joint dependency on Oracle's cloud operations and HPE's networking supply chain. Model dual-vendor outage scenarios explicitly, and require contractual clarity on which party owns incident response for network-layer failures in OCI regions.